Revisiting Self-Organizing Maps for Drug-Disease Association Prediction: A Graph-Based Approach

ICLR 2025 Workshop LMRL Submission69 Authors

Track: Tiny Paper Track

Keywords: Graph Neural Networks, Self-Organizing Maps, Drug-Disease Association Prediction

TL;DR: We integrate Graph Neural Networks and Self-Organizing Maps to enhance drug-disease association prediction, demonstrating SOMs' potential in biomedical informatics.

Abstract: We present an advanced framework integrating Graph Neural Networks (GNNs) and Self-Organizing Maps (SOMs) for drug-disease association prediction. While GNNs efficiently model structured biomedical data, SOMs remain underexplored despite their strong unsupervised clustering capabilities. Our approach constructs a heterogeneous graph, employs GraphSAGE for link prediction, and applies SOMs for embedding space organization. The results reveal that SOMs effectively capture latent relationships, reinforcing their potential for biomedical informatics. This work advocates for further exploration of SOMs as a complementary tool in computational drug discovery.

Submission Number: 69